KNC ASIC Users Thread & FAQ

[nwoolls]

According to this post on the KnC miners forum, there are users interested in using MultiMiner with their KnC miners.

Any of you lucky folks feel like testing the Stratum Proxy feature in MultiMiner to see if it works with your KnC miner? It currently works well with Erupter Blades and I'm curious to see if it will work with KnC miners.

First, download and run MultiMiner. Go into the Settings and choose bfgminer as the backend miner. Then click Advanced backend miner settings and enable the Stratum Proxy setting.



This will cause a "virtual" device to show in MultiMiner which you can then use to configure coins, pools, failover, etc.

Then you should be able to configure your KnC miner to use the IP address and port of the Stratum Proxy configured above.

Again it works with the Blade miners and I am curious to see if it works with the KnC hardware as well. Thanks in advance if anyone feels like having a play and leaving some feedback!

[ecliptic]

Suggestion: Figure out how to mine through Tor or similar anonymizing network

Remember, several people with early batch Avalons had their homes broken into and their units stolen.

Why take unnecessary risk?

Yeah mining through tor...no latency or connectivity issues there.  How about just not posting your IP, address, or order information....otherwise known as common sense.

The pool has your IP.  That is more than enough.

God help you if you use P2Pool.

[gnar1ta$]

Suggestion: Figure out how to mine through Tor or similar anonymizing network

Remember, several people with early batch Avalons had their homes broken into and their units stolen.

Why take unnecessary risk?

Yeah mining through tor...no latency or connectivity issues there.  How about just not posting your IP, address, or order information....otherwise known as common sense.

The pool has your IP.  That is more than enough.

God help you if you use P2Pool.

Oh no...all my Jupiter has been stolen!!!  I shouldn't have used p2pool.... Oh wait, back to reality, everything is good...still hashing.

[fpgaminer]

Reporting my experience so far:

I have one of the duds.  v0.9.4, case off, box fan.  Only 440 GH/s, ~16% error rate, and 900W.

Using Bertmod 0.2 I was able to look at the voltages and such.  Looks like my unit is being overvolted to 0.92V on all the modules.  As far as I can tell the nominal voltage is 0.7V, so I think their firmware is choosing to overvolt to compensate for bad chip(s).  So all the dies are pulling 50 Amps, driving my total power usage to 900W.

I suspect that not all of the dies are bad; only some.  But the firmware is indifferent to the particular needs of each die, and simply overvolts all of them to keep up the hashrate.  If that's the case, I could see a future update which only overvolts the under-performing dies, reducing my overall power consumption.

It's better than nothing, and I sincerely understand that KNC's schedule was extraordinarily tight.  I commend them for meeting their goals on such an ambitious project.  It's still disappointing to see some people getting good units, and others getting bad units, though

EDIT: I poked around with the firmware.  It seems that I was wrong about it setting the voltage to 0.9V automatically.  Rather, it looks like it's that way from the factory; the firmware isn't setting the voltage.  Though the screenshot here shows someone at 0.7V.  So for whatever reason KNC is shipping some units at 0.7V and some at 0.9V.

On top of that, unless the DC/DC modules they're using have changed, these are only rated for 40A.  Mine are pushing 50A.  ...

I did discover, though, that it's possible to digitally tweak the voltage.  Perhaps a future firmware upgrade will take advantage of that.  I don't have the balls to try it myself manually.

[warhawk187]

Does anyone know why my Saturn stops hashing after 15 minutes???! This is ridiculous, it will hash @275-280 and then cgminer will start throwing FPGA errors and then either die completely or turn off a core and run @170ish Gh/s, but its never consistent. I've used 3 different PSU's and used all of the available firmware versions (mine came with 0.91).

What the F??! Anyone have any insight, or at least know how I can script cgminer via Putty to restart every 10 minutes? The case is off and when it hashes it hashes fine, but then, it just....dies. What gives?!?!

[hagbase]

Try this:
downgrade to 0.90 and upgrade than to 0.94, do a hard reset, wait till miner is online, maybe u'll get an error message pool config is wrong or something like that, connect with putty and restart cgminer/config.

I've mounted two Noctua NF-S12B FLX Fans sucking air out to streamline the airflow in case.

[warhawk187]

Try this:
downgrade to 0.90 and upgrade than to 0.94, do a hard reset, wait till miner is online, maybe u'll get an error message pool config is wrong or something like that, connect with putty and restart cgminer/config.

I've mounted two Noctua NF-S12B FLX Fans sucking air out to streamline the airflow in case.

Thank you! The good news is that has been stable for at least 20 minutes now. The bad news is that it now runs at ~230Gh/s instead of ~275Gh/s for the 10 minutes it would run before throwing endless FPGA core 21 errors. I hope they sort this out soon.

[billywilly]

Someone knows what does it mean "FAULT 17" on status of VRM's?


https://i.imgur.com/c8tDprM.png

[sickpig]

Someone knows what does it mean "FAULT 17" on status of VRM's?

never seen before. how many cores are enabled for that die/asic?

[sickpig]

Someone knows what does it mean "FAULT 17" on status of VRM's?

never seen before. how many cores are enabled for that die/asic?

fw 0.95 was just released. give it a try.

[fpgaminer]

Just updated to 0.95; seems to have fixed my problems.  Went from 440GH/s, 16% errors, 900W to 530GH/s, 5% errors, 620W.  Much better!

As I suspected, this update dropped my voltages from 0.92V to ~0.73V, thus the drop in power usage.  Obviously they must have also fixed something else that corrected whatever hashrate issues units like mine were having.

[ralree]

Just got my jupiter and installed 0.95.  After 30 minutes or so of running and slowly creeping to 420GH or so, cgminer restarted and the unit now is averaging 475GH or so, with spikes above 500 being rather common.  Lots of the cores are going off and on due to hardware errors over and over again.  Overall, I'm pretty happy, but I wish it was over 500GH consistently.  It's pulling 2A @ 220V from my PDU, so I'm assuming it's rounding down and it's just above 500W.

[mininganon]

Upgraded from .91 to .95

Erratic first hour, it would go up to 550ghs then crash to 300 something and back. It looks to be settling now at about 450ghs. Will update

[mininganon]

.95 firmware for me is a little slower but more stable... Really torn here between loading .91 back and keeping .95?

Don't know how KNC is going to fix my issue if they are shipping 4vrm per module now and I have 8. It's like I have a different product.

Not always best to be first!

[ecliptic]

.95 nets me around 275/257 GH/sec for a Saturn on cgminer

the crazy thing is the onboard temps don't read above 48C, and i'm not seeing ANYTHING - VRM modules, caps, etc above 55C at the most.

Makes me think they're not thermally limited for clock rate.  Maybe the buck converters can't push enough current to clock higher or there's not sufficient decoupling capacitance/other voltage issues?

Or do these have untapped potential?  I am ignorant to how these devices are clocked, if cgminer / etc software can adjust the clock frequencies or if it's a hard-soldered oscillator / clock divider or what.

301 watts at the PSU line in

this is also with the cover off, but no external fans

[edgar]

.95 firmware for me is a little slower but more stable... Really torn here between loading .91 back and keeping .95?

Don't know how KNC is going to fix my issue if they are shipping 4vrm per module now and I have 8. It's like I have a different product.

Not always best to be first!

also very interested in the 8 vs 4 VRM boards....

seems to be the difference between a stable 550 and an unstable 400+

[jelin1984]

Which is stable and which is unstable ??with 0.95

The 4 or 8 version?

[billywilly]

Someone knows what does it mean "FAULT 17" on status of VRM's?


https://i.imgur.com/c8tDprM.png

never seen before. how many cores are enabled for that die/asic?

Now upgraded from 0.90 to 0.95 and the errors went off, but the cores disabled on that core were 48 in total, that's the reason because that core ran much cooler than the others. Dropped to 430w.

[Tigggger]

Added details of how to show the correct times into the first post

Modifications
Fixing the time.
Machines appear to ship with the time set to UTC, although it's not important it did annoy me not having the correct time, these instructions came from Nanners on the IRC channel and worked for me, replace 'tigggger' with whatever you want.

1. cd /config
2. mkdir tigggger
3. cd /config/tigggger
4. wget http://www.angstrom-distribution.org/feeds/v2012.12/ipk/eglibc/all/tzdata_2012d-r3.0_all.ipk
5. opkg install /config/tigggger/tzdata_2012d-r3.0_all.ipk
6. rm /etc/localtime (NOTE this gave an error for me as the directory wasn't there, but not important just carry on)
7. ln -s /usr/share/zoneinfo/YOURTIMEZONE /etc/localtime (In my case it was /usr/share/zoneinfo/Europe/Belfast)
8. cd /
9. /etc/init.d/ntpd reload
10. ntpdate -b -u pool.ntp.org

Now if you look at cgminer the time should be correct.

[btc_uzr]

if you want to change the timezone only, you can set your env var 'TZ=GMT'  (or whatever you prefer)
https://bitcointalk.org/index.php?topic=170332.msg3337684#msg3337684

However it will not sync your time with ntp servers as the approach above from Tiggger